AI Documents

Digital Transformation - Archives

Ross Dawson writes:

For all the talk of AI Agents there is little clarity. Over the next while I'll share a number of frameworks to lay out different aspects of AI agents, what they are, how they function, and the impact on business models.

One of the important distinctions is in their level of functionality, from simple rule-based execution, through to self-learning orchestrators.

Here is the first

The slides describe each level, including characteristics, technologies, use cases, and insights. Here they are in short, from the simplest to the most powerful.

🔧 Rule-Based Executors Follows predefined rules to execute structured, repetitive tasks without adaptability or intelligence.

📚 Knowledge Augmented Agents Grounds decisions in real-time external data and domain-specific knowledge for precision-critical tasks.

🔄 Integrated Workflow Executors Manages complex workflows by dynamically integrating diverse systems and tools seamlessly.

🤔 Explainable Decision Agents Meta-cognitive agents that can explain their reasoning, refine decisions, and learn from feedback.

👤 Personalized Task Handlers Maintains historical context to provide consistent, user-specific interactions and recommendations.

🌐 Real-Time Environment Managers Controls physical or digital environments dynamically using IoT and robotics systems with real-time feedback.

🧠 Self-Learning Orchestrators Autonomously learns, adapts, and improves over time while coordinating and evolving complex processes.

Following frameworks will look at different frames around definitions, applications, and business models.

Om N. writes:

99% people on LinkedIn still think that “Chatbots” are AI Agents

AI Agent is becoming the new buzzword on LinkedIn and people often use it interchangeably with Automation or AI Workflow - But they’re poles apart!

What's the difference?
→ Automations execute predefined, rule-based tasks automatically
→ AI Workflows are automations that call LLMs like ChatGPT via API for one or more steps
→ AI Agents are programs designed to perform non-deterministic tasks autonomously

Which tasks can they best handle?
→ Automations shine with pre-defined deterministic tasks
→ AI Workflows are great for deterministic tasks requiring some flexibility
→ AI Agents should be used to handle non-deterministic, adaptive tasks

What are their strengths?
→ Automations deliver outcomes reliably and are fast to execute
→ AI Workflows are great for pattern recognition and handling complex rules
→ AI Agents are best when you expect new variables and scenarios

What are their respective weaknesses?
→ Automations limited to tasks explicitly programmed and cannot adapt
→AI Workflows require data to train models and are usually harder to debug
→AI Agents are less reliable and may produce unpredictable outcomes

If you’re building AI Agents, checkout my profile for resources 👋

credits : Alexandre Kantjas

Impressions by Winnie Ngige on The Near-Term Impact of AI on Disinformation by Tommy Shaffer Shane

My reading this weekend has been the Article 'The Near-Term Impact of AI on Disinformation'. This article offers a detailed exploration of how AI is transforming the disinformation landscape.

First, it begins by categorising threat actors. We have the Jokers, who are individuals with AI expertise who create satirical or entertaining content. Their intent is often lighthearted, but their content can easily be stripped of its context and repurposed for harmful purposes. Highlighting the unintended consequences of seemingly harmless AI applications.

Next are the lone rogrammers. These are technically skilled individuals who exploit open-source AI models to develop programs. Their creations, while not inherently malicious, often become tools in the hands of actors seeking to spread disinformation.

We also have domestic networks, these are organisations such as political parties which use AI-driven disinformation to further political ambitions. These actors leverage the technology to manipulate public opinion during elections or other politically charged moments, presenting a direct threat to democratic processes.

Following closely are capable states and Firms. According to the article, these entities, driven by financial or political objectives, often outsource their disinformation efforts to professional firms offering "disinformation for hire." These firms craft tailored narratives that manipulate social and cultural vulnerabilities, amplifying their reach and impact.

At the apex of this ecosystem are highly capable states. These ones are powerful actors with vast resources and expertise. They run large-scale disinformation campaigns aimed at quelling dissent or undermining adversaries. The article emphasizes their significant role in the spread of disinformation, given their ability to execute complex and far-reaching operations.

The article further points out the impact of AI in different populations and audiences. For instance, it highlights the capacity of generative AI in not only changes who can participate in disinformation but also how it spreads.

It also presents the troubling statistic that 90% of deepfake content targets women, a trend expected to worsen as AI tools become more sophisticated.

The article further examines the role of AI in influencing information and democracy. It states that beyond the mechanics of creation, AI-driven disinformation thrives by exploiting societal vulnerabilities. By capitalizing on existing divisions, it fosters confusion and distrust, eroding public confidence in reliable information. This has far-reaching implications for social cohesion and institutional credibility.

To address the identified issues, the article proposes the following:

📌International standards citing that global cooperation is critical in enabling ethical AI development.

📌Tools to monitor the evolving threat labdscape to identify and mitigate emerging risks.

Happy reading 😊

Martha Njeri's Impressions on Emerging Practices for AI Impact Assessments, a report by Future of Privacy Forum

AI Governance Behind the Scenes - Emerging Practices for AI Impact Assessments

This report by Future Privacy Forum digs into the evolving landscape of AI impact assessments, which are becoming increasingly crucial for organizations seeking to manage the risks associated with AI technologies. The report highlights the key steps involved in these assessments, the challenges organizations face, and emerging trends in the field, drawing on research and input from over 60 companies.

It is also worth noting that companies are increasingly adopting AI in their operations and systems. For example, in the banking industry, AI is widely used in credit scoring and fraud detection (human oversight is always recommended). The healthcare industry has not been left behind, with AI being widely used in optimizing clinical workflows and the rise of robotic surgeries. And last but not least, the retail industry also tops the list in the use of AI for personalized shopping recommendations, etc. With the growing influence of AI across sectors, conducting thorough AI Impact Assessments becomes increasingly important.

This report highlights the following key aspects of the AI impact assessment namely:
(a)Initiating an AI Impact Assessment:
Where various factors can trigger an AI impact assessment, such as legal requirements, governance norms, and ethical or business risks.
(b)Gathering Model and System Information:
Organizations should gather information to understand risks and benefits, focusing on data.
(c)Assessing Risks and Benefits:
This involves analyzing the types and levels of risk, potential benefits, and the technical and legal context.
(d)Identifying and Testing Risk Management Strategies:
This involves organizations selecting strategies based on their responsiveness to specific identified risks.

Challenges mentioned include:
-Organizations struggle to obtain information from third-party model developers and system vendors.
-Anticipating all pertinent AI risks is difficult.
-Determining whether risks have been brought within acceptable levels is challenging.

The report recommends enhancing processes for gathering information, improving internal education about AI risks, and devising better measurements for risk management strategies. Additionally, AI governance training across the organization and executive-level support are essential.

Summarily, the report underscores that AI impact assessments are crucial for responsible AI development and use, emphasizing the need for continuous improvement and adaptation in this rapidly evolving field.

Sam Burret speaks to the rise of "AI Overconfidence."

He says new research finds:

- 96% of employees 'trust AI for complex work'.
- But 58% of leaders say they struggle to demonstrate ROI from AI.

Employees are overly trusting in AI; organisations increasingly skeptical. And it's often because of 3 things:

1/ Unchecked experimentation
Without structured programs, employees learn about AI through experimentation. That's great - but can easily lead to overconfidence. Especially for junior professionals who are less able to critique AI outputs.

2/ Not solving a business problem
When AI use isn't tied to specific business problems, we get hyped about the tech itself - rather than the impact. It becomes too easy to trust outputs, with no-one asking, "is this really doing anything meaningful?"

3/Lack of scaling frameworks
Employees see immediate personal productivity gains from AI. That creates trust. But without parameters or metrics, we can't translate this into business value. Hence the trust-ROI disconnect.

These are very common - but very solvable problems.

Our AI Advisory Practice is working with clients to ensure they have the programs and processes to ensure AI adoption is scalable, strategic, and responsible.

Confidence in AI is great. But a strategy to turn that into value - that's way better 😉

Link for more information: https://www.linkedin.com/feed/update/urn:li:activity:7287732033659850753/